Guided Sampling Method for Hypothesis Generation in Image Stitching

The paper is focused on the study of the guided sampling method for estimating homographies in the problem of image stitching. Traditionally, to solve this problem, the procedure of sequential partitioning of points into subclasses belonging to different planes or estimation of local transformations is used. In this paper we consider the solution of this problem by the method of guided sampling. Experimental study of the influence of the algorithm parameters on the quality of the obtained estimates on synthetic and real data is carried out. It is shown that the joint use of the guided sampling method and the proposed clustering algorithm allows to reliably divide the points of the initial set into subsets containing points located on different planes of the scene. This separation ensures the determination of different homographies for the corresponding points in the images.